Semantic lattices for multiple annotation of images

Anne-Marie Tousch (1, 2, 3, 4) , Stéphane Herbin (1) , Jean-Yves Audibert (2, 3, 4, 5)

1 ONERA - The French Aerospace Lab [Palaiseau]
2 CERTIS - Centre d'Enseignement et de Recherche en Technologies de l'Information et Systèmes
3 imagine [Marne-la-Vallée]
4 LIGM - Laboratoire d'Informatique Gaspard-Monge
5 WILLOW - Models of visual object recognition and scene understanding

We address the problem of describing precisely an object present in an image. The starting point is a semantic lattice defining all possible coherent object descriptions through inheritance and exclusion relations. This domain knowledge is used in a learning process which outputs a set of coherent explanations of the image valued by their confidence level. Our first contribution is to design this method for multiple complexity level image description. Our secondary focus is to develop rigorous evaluation standards for this computer vision task which, to our knowledge, has not been addressed in the literature despite its possible use in symbolic annotation of multimedia database. A critical evaluation of our approach under the proposed standards is presented on a new appropriate car database that we have collected.

Domaines: Informatique [cs]; Vision par ordinateur et reconnaissance de formes [cs.CV]

Anne-Marie Tousch, Stéphane Herbin, Jean-Yves Audibert. Semantic lattices for multiple annotation of images. 1st ACM international conference on Multimedia information retrieval, Oct 2008, Vancouver, Canada. pp.342-349, ⟨10.1145/1460096.1460152⟩. ⟨hal-00835102⟩